Background: The clinical significance of pulmonary nodular ground-glass opacities (NGGOs) in patients with extrapulmonary cancers is not known, although there is an urgent need for study on this topic. The purpose of this study, therefore, was to investigate the clinical significance of pulmonary NGGOs in these patients, and to develop a computerized scheme to distinguish malignant from benign NGGOs.
Methods: Fifty-nine pathologically proven pulmonary NGGOs in 34 patients with a history of extrapulmonary cancer were studied. We reviewed the CT scan characteristics of NGGOs and the clinical features of these patients. Artificial neural networks (ANNs) were constructed and tested as a classifier distinguishing malignant from benign NGGOs. The performance of ANNs was evaluated with receiver operating characteristic analysis.
Results: Twenty-eight patients (82.4%) were determined to have malignancies. Forty NGGOs (67.8%) were diagnosed as malignancies (adenocarcinomas, 24; bronchioloalveolar carcinomas, 16). Among the rest of the NGGOs, 14 were atypical adenomatous hyperplasias, 4 were focal fibrosis, and 1 was an inflammatory nodule. There were no cases of metastasis appearing as NGGOs. Between malignant and benign NGGOs, there were significant differences in lesion size; the presence of internal solid portion; the size and proportion of the internal solid portion; the lesion margin; and the presence of bubble lucency, air bronchogram, or pleural retraction (p < 0.05). Using these characteristics, ANNs showed excellent accuracy (z value, 0.973) in discriminating malignant from benign NGGOs.
Conclusions: Pulmonary NGGOs in patients with extrapulmonary cancers tend to have high malignancy rates and are very often primary lung cancers. ANNs might be a useful tool in distinguishing malignant from benign NGGOs.